Examining the impact of the Great Barrier Reef on tsunami propagation using numerical simulations

Abstract

Coral reefs may provide a beneficial first line of defence against tsunami hazards, though this is currently debated. Using a fully nonlinear, Boussinesq propagation model, we examine the buffering capacity of the Great Barrier Reef against tsunamis triggered by several hypothetical sources: a series of far-field, Solomon Islands earthquake sources of various magnitudes (M_w 8.0, M_w 8.5, and M_w 9.0), a submarine landslide source that has previously been documented in the offshore geological record (the “Gloria Knolls Slide”), and a potential future landslide source (the “Noggin Block”). We show that overall, the Great Barrier Reef acts as a large-scale regional buffer due to the roughness of coral cover and the complex bathymetric features (i.e. platforms, shoals, terraces, etc.) that corals construct over thousands of years. However, the buffering effect of coral cover is much stronger for tsunamis that are higher in amplitude. When coral cover is removed, the largest earthquake scenario (M_w 9.0) exhibits up to a 31% increase in offshore wave amplitude and estimated run-up. These metrics increase even more for the higher-amplitude landslide scenarios, where they tend to double. These discrepancies can be explained by the higher bed particle velocities incited by higher-amplitude waves, which leads to greater frictional dissipation at a seabed covered by coral. At a site-specific level, shoreline orientation relative to the reef platforms also determines the degree of protectiveness against both types of tsunamis, where areas situated behind broad, shallow, coral-covered platforms benefit the most. Additionally, we find that the platforms, rather than gaps in the offshore reef structure, tend to amplify wave trains through wave focussing when coral cover is removed from simulations. Our findings have implications for future tsunami hazards along the northeastern Australian coastline, particularly as the physiological stressors imposed by anthropogenic climate change further exacerbate coral die-off and reductions in ecosystem complexity. Therefore, areas that experience a protective benefit by the Great Barrier Reef’s platforms could be disproportionately more vulnerable in the future.

1 Introduction

Tsunamis threaten low-lying coastal communities around the world. Coral reef ecosystems, many of which are positioned between tsunami source regions and densely populated shorelines (Fig. 1), could provide a broad, cost-effective first line of defence for coastal zones (Ferrario et al. 2014). While field-based studies suggest that coral reefs induce efficient energy attenuation in wind waves due to their structural complexity (Sheppard et al. 2005; Ferrario et al. 2014; Gallop et al. 2014), a lack of consensus endures surrounding their protectiveness against tsunamis.

Fig. 1

source zones of major tsunamigenic earthquakes. Landslides are plotted as red circles sized proportionally to the natural log of a given landslide’s volume. This compilation is based on several reviews (Hampton et al. 1996; Elverhøi et al. 2002; Owen et al. 2007; Lee 2009; Urlaub et al. 2013; Harbitz et al. 2014; Papadopoulos et al. 2014; Moscardelli and Wood 2016), where landslides with estimated volumes of 1 km³ were excluded. All original references documenting each of the plotted slides are provided in the reference list of this study. Landmasses are overlaid with gridded UN-adjusted population density for 2020 (CIESIN 2018), with ETOPO1 as the base map (Amante and Eakins 2009)

Global distribution of shallow-water coral reefs (Burke et al. 2011) and their proximity to tsunamigenic sources, including large submarine landslides or landslide complexes (> 1 km³; see Online Resource 1 for table of landslide events) and submarine convergent plate boundaries that constitute

Following a similar logic, some post-inundation field surveys (Fernando et al. 2005; McAdoo et al. 2011) and modelling studies (Shao et al. 2019) have concluded that, due to their structural complexity, coral reef ecosystems impart similar drag-induced attenuation of wave energy on tsunamis. Other field-based studies (McAdoo et al. 2009; Fritz et al. 2011; Gelfenbaum et al. 2011) and modelling work (Kunkel et al. 2006; Yao et al. 2012; Roger et al. 2014) echo these conclusions, but with caveats. For instance, some authors caution that the buffering effect of the reef depends on where the reef is located relative to a coastal community or built asset (McAdoo et al. 2009; Fritz et al. 2011), and that wider reefs, preferably those with an extensive reef flat, appear to dissipate tsunami energy more effectively than narrower fringing reefs (Kunkel et al. 2006; Gelfenbaum et al. 2011; Yao et al. 2012; Roger et al. 2014). Conversely, others have proposed that coral reefs offer marginal to no protective benefit against tsunamis (Baird et al. 2005; Uslu et al. 2010). Further still, some field-based studies (Nott 1997; Chatenoux and Peduzzi 2005, 2007; Fritz et al. 2011) and modelling work (Roeber et al. 2010; Gelfenbaum et al. 2011; Yao et al. 2012; Ford et al. 2014) suggest that reefs can actually exacerbate damage along neighbouring coastlines. While there is near-universal consensus that inter-reef passages (or “gaps/openings” between reefs) can amplify tsunami waves, some argue that these amplification effects, along with other effects such as intra-lagoon resonance and increased shoaling and bore formation over shallow reef platforms, undermine any protective benefit that the presence of the reef would otherwise offer (Chatenoux and Peduzzi 2005; Liu et al. 2005; Roeber et al. 2010; Gelfenbaum et al. 2011; McAdoo et al. 2011; Ford et al. 2014; Roger et al. 2014). Despite the wide variety of methods and case studies employed to investigate this topic, the impact of coral reef ecosystems on tsunami propagation remains equivocal.

Ongoing threats to the health and longevity of coral reefs under a changing climate (De’ath et al. 2012; Hughes et al. 2018) heighten these uncertainties. Decades-long field-based studies reveal declines in both coral cover and ecosystem structural complexity as critical reef-building species disappear from coral communities, leading to a progressive “flattening” of reefs (Alvarez-Filip et al. 2009; Bozec et al. 2015; Spalding and Brown 2015). It has been proposed that this decline in coral cover will reduce the protectiveness of coral reefs against other common coastal hazards, such as flooding, wind-wave exposure (both under fair weather and stormy conditions), and rising sea levels (Quataert et al. 2015; Harris et al. 2018; Storlazzi et al. 2018). The literature surrounding the impact of anthropogenically mediated coral decline on tsunami hazards is less conclusive. However, some evidence from post-tsunami field surveys suggests that direct coral removal by means of mining and poaching intensifies tsunami wave heights and inundation extents at a local level (Fernando et al. 2005). In light of recent coral reef decline, and in the wake of recent significant tsunami events (e.g., the 2004 Indian Ocean tsunami, the 2009 South Pacific tsunami, and the 2011 Tōhoku tsunami), a concerted effort has emerged to more rigorously assess both the present and future coastal buffering role of coral reef ecosystems against tsunamis (Chatenoux and Peduzzi 2007; Ferrario et al. 2014; Spalding et al. 2014), and this study is a contribution to that effort.

The Great Barrier Reef (GBR), the world’s largest coral reef system, is an iconic feature of Australia’s coastal landscape. Despite Australia’s proximity to seismically active source regions (Dominey-Howes 2007; Davies and Griffin 2018), the manner in which tsunami behaviour is regulated by the GBR, which partitions Australia’s coastline from these convergent margins, is not well understood (Webster et al. 2016). Additionally, the discovery of large (volume > 30 km³) landslide scars and slumps on the nearby continental slope (Puga-Bernabéu et al. 2016, 2019) warrants an investigation into the GBR’s ability to protect against landslide-generated tsunamis. Though believed to occur less frequently than their coseismic counterparts, landslide-generated tsunamis such as the 1998 Sissano, Papua New Guinea event (Synolakis et al. 2002) can occur suddenly within close proximity to the shoreline, causing significant localized damage and limiting opportunities for warning and swift response. This, along with the existence of possible paleo-tsunami deposits along the adjacent coastline (Nott 1997), underscores an urgency to quantify the GBR’s widely speculated role as a regional buffer from these hazards (Baba et al. 2008; Puga-Bernabéu et al. 2013a; Wei et al. 2015; Xing et al. 2015; Webster et al. 2016). However, like most coral reefs worldwide, the GBR has not escaped the consequences of anthropogenic climate change (De’ath et al. 2012; Hughes et al. 2018), and therefore, the future buffering capacity of the GBR remains uncertain.

Thus far, a large portion of the debate surrounding coral reef protectiveness against tsunamis is based on findings from post-tsunami field surveys and anecdotal eye-witness accounts (Baird et al. 2005; Fernando et al. 2005; Liu et al. 2005). However, the degree of a coral reef’s influence cannot be quantified solely from these field-based techniques. As many others have highlighted (Chatenoux and Peduzzi 2005; Kunkel et al. 2006; McAdoo et al. 2009; Uslu et al. 2010; Roger et al. 2014; Dilmen et al. 2018), several confounding factors can influence tsunami run-up, such as the extent of coral cover, the nature and proximity of the tsunami triggering source, and site-specific variability in coastal bathymetry and topography. Therefore, following a tsunami event, it is difficult to retrospectively ascertain the impact of coral reefs in isolation from these other site-specific factors. Numerical simulations can provide additional insights into tsunami behaviour (e.g., Kunkel et al. 2006), where experiments can be designed to systematically test the impact of coral cover and reef platform bathymetry on tsunami attenuation while keeping all other parameters, initial conditions, and boundary conditions constant (e.g., Kunkel et al. 2006). Previous studies have aimed to assess the overall impact of the GBR on tsunami propagation using numerical simulations (Baba et al. 2008; Wei et al. 2015; Xing et al. 2015; Webster et al. 2016). However, they do not account for smaller-scale structural complexity introduced by coral cover on reef platforms, and they only consider one type of tsunami source at a time.

Using numerical modelling, we evaluate the GBR’s ability to shield the northeastern Australian coastline from a range of hypothetical, though plausible tsunami sources. Firstly, we consider a Solomon Islands earthquake source over various magnitudes (M_w 8.0, M_w 8.5, and M_w 9.0). Additionally, we consider two near-field landslide tsunami sources: (1) the largest documented submarine landslide event on the GBR margin (i.e. the Gloria Knolls landslide complex; Puga-Bernabéu et al. 2016), and (2) a potential collapse of a feature on the upper continental slope known as the Noggin Block (Puga-Bernabéu et al. 2013a).

In the first of a series of tsunami propagation model runs, for each tsunami source, we numerically simulate the tsunamis assuming healthy coral cover conditions (i.e. “coral-covered platforms” scenarios), where reef platforms are prescribed high roughness to reflect their structural complexity (Nelson 1996). Then, we simulate the tsunamis with smoothed reef platforms (i.e. “smooth platforms” scenarios), where we isolate the impact of live coral cover on wave attenuation (Sheppard et al. 2005). Following the methods of Baba et al. (2008), we further sequester the region’s bathymetric complexity by completely excising the reef platforms from the shelf and simulating tsunami propagation with altered bathymetry (i.e. “no reef platforms” scenarios), allowing us to assess the platform-scale buffering capacity of the entire reef structure. We further test the impact of tidal phase on the buffering capacity of the GBR. We then draw upon these findings to consider the broader implications regarding present and future coral reef defence to densely inhabited, low-lying coastal areas.

2 Study area

2.1 Regional setting

The central northeastern Australian margin is a passive margin characterised by a relatively broad (~ 60 km) continental shelf (Fig. 2). The spring tidal range varies from north to south, but the region is generally meso- to macro-tidal (Andrews and Bode 1988). Several environmental factors favour coral reef growth on the mid- to outer continental shelf, including the region’s tropical climate, shallow seas, far proximity from terrestrial run-off, and nutrient-poor oceanographic conditions. Over hundreds of thousands of years of eustatic sea level fluctuations, these coral reef ecosystems have constructed large (up to ~ 300 km²) submerged and semi-submerged carbonate platforms, pinnacles, and terraces, which comprise the offshore reef structure (Hopley et al. 2007; Hinestrosa et al. 2016). This reef structure, which underlies the modern generation of living coral cover, extends roughly 2,300 km along the mid- to outer shelf (Hopley et al. 2007). On the central margin, broad, arcuate patch reef platforms are separated by relatively wide (up to ~ 10 km) inter-reef passages, or “gaps” (Fig. 3a). While these passages are wide enough to allow some wind waves to propagate through to the inner shelf, much of the energy transferred by wind waves is attenuated atop the reef platforms (Young 1989; Gallop et al. 2014).

Fig. 2

source zone, the Coral Sea, and the northeastern Australian margin, which includes the GBR (orange). Also plotted are the locations along the Australian coastline where historical tsunamis that exceeded maximum water heights of 10 cm have been observed using tide gauges (triangles; NGDC/WDS 2020). The red line indicates the subduction zones that traverse the Solomon Islands source zone

Regional view of the Solomon Islands

Fig. 3

a Bathymetry used in the “coral-covered platforms” and “smooth platforms” simulations. b Bathymetry used in the “no reef platforms” simulations. Also shown are the Gloria Knolls Slide, the Noggin Block, ODP Leg 133 Site 819, the locations of the boulder deposits described by Nott (1997) and an inset showing the location of the hypothetical Solomon Islands coseismic sources

2.2 Historical and pre-historic tsunami record

Historically, northeastern Australia has been affected by tsunamis originating from multiple regions contained within the Pacific Ring of Fire (e.g. Chile, Tonga, and more recently, Sumatra and Japan; see Fig. 2). Notably, a large proportion of these historical tsunami events were triggered within subduction zones in the Solomon Islands region, which lies to the northeast of Australia across the Coral Sea (Dominey-Howes 2007; Australian Bureau of Meteorology 2020; NGDC/WDS 2020). A nationwide, probabilistic tsunami hazard assessment revealed that the Solomon Islands source zone poses the greatest hazard to the northeastern Australian town of Cairns and the surrounding area (Davies and Griffin 2018). Therefore, the Solomon Islands source zone was selected to simulate a range of hypothetical earthquake-generated tsunami events for this study. In contrast, the pre-historic tsunami record in northeastern Australia is much more sparse (Dominey-Howes 2007). Nonetheless, previous work has described boulder deposits that were speculated to have been emplaced by tsunami waves (Nott 1997; Fig. 3).

2.3 Submarine landslides and areas of potential future collapse

Since the collection of high-resolution multibeam bathymetry in 2007 (Webster et al. 2008), a wide variety of submarine landslides have been described on the shelf-edge, upper, mid-, and lower slope (Puga-Bernabéu et al. 2016, 2019; Webster et al. 2016). These slides exhibit a range of different sizes and morphologies (e.g. rotational slumps, translational slides, shovel slides, carbonate terrace collapses, etc.). While they are distributed along the entirety of the margin, landslides are more commonly found on the north and central sections of the margin, where the continental slope gradient is moderate to high (4°–10°, Puga-Bernabéu et al. 2011, 2013b).

The present study focuses on two notable features on the central GBR margin. The first is the Gloria Knolls landslide complex (Puga-Bernabéu et al. 2016), which is the largest amongst the documented submarine landslide cases on the northeastern Australian margin (total estimated volume ≈ 32 km³). The entire complex is believed to have failed in multiple phases, with the estimated age of the first event pre-dating 300 ka (Puga-Bernabéu et al. 2016). Debris from the slide is visible in both sub-bottom profiles and in bathymetry, where the debris field extends ~ 20 km from the slide scarp. Roughly 8 km northwest of the Gloria Knolls slide complex lies the Noggin Block, a 4.9 × 3.5 km upper-slope feature that was previously identified as a potential area of future collapse (Puga-Bernabéu et al. 2013a). Pockmarks and adjacent landslide scars have also been described around the block (Puga-Bernabéu et al. 2013a). Slope stability modelling indicates that while the block is presently stable, seismic loading could potentially trigger a future failure (Puga-Bernabéu et al. 2013a).

We should note that it lies beyond the scope of this work to include a detailed catalogue, and thus a detailed hazard assessment, of landslide tsunami risk on this margin. A complete catalogue of all submarine landslides on the GBR margin is currently the subject of future work (Puga-Bernabéu et al., in prep).

3 Methods

3.1 Tsunami generation

3.1.1 Earthquake sources

To simulate tsunami generation by an earthquake source, the code Geowave (Watts et al. 2003) was used to produce the initial ocean free surface deformation for the hypothetical M_w 8.0, 8.5, and 9.0 coseismic events in the Solomon Islands source zone. Tsunami generation is specifically handled in the TOPICS module of Geowave (Watts et al. 2003). The code incorporates the widely implemented Okada elastic half-space formulation, which relates earthquake geometric source parameters (i.e. fault width, length, strike, dip, etc.) to the initial free surface deformation (Okada 1985). The Okada method has been shown to adequately reproduce free surface deformation for coseismic events exhibiting an abrupt, mostly vertical slip of the seafloor (Kowalik et al. 2005; Fujii et al. 2011) and specifically for past events that originated in the Solomon Islands (Baba et al. 2008). Source parameters were selected from the Enhanced Tsunami Scenario Database T2 (Greenslade et al., 2009; see Table 1), a suite of earthquake tsunami scenarios developed by the Joint Australian Tsunami Warning Centre and the Centre for Australian Weather and Climate Research. For simplicity, magnitude was altered by modifying the maximum fault slip parameter (see Table 1).

Table 1 List of input parameters used for tsunami wave generation models.

3.1.2 Submarine landslide sources

To simulate tsunami generation by the Gloria Knolls Slide and the potential collapse of the Noggin Block, we used NHWAVE (Ma et al. 2013), a non-hydrostatic wave model that has been successfully validated in laboratory settings (Enet and Grilli 2007; Tehranirad et al. 2012) and has been used for several case studies of submarine mass failure-induced tsunamis (Tappin et al. 2014; Grilli et al. 2015; Li et al. 2015; Schnyder et al. 2016). The code numerically approximates the solutions to the non-hydrostatic Navier–Stokes equations for incompressible flow in three dimensions, implementing a terrain-following (i.e. sigma-layered) vertical coordinate system. For simplicity and computational efficiency, a three-dimensional, rigid, translational failure was assumed for both cases, where the bottom boundary condition is dictated by a time-varying change in depth imparted by an approximately Gaussian-shaped slide.

NHWAVE requires approximate landslide dimensions (i.e., length, width, thickness) to construct the Gaussian-shaped slide that generates the initial tsunami. For both landslide cases, these dimensions were determined in previous work (Puga-Bernabéu et al. 2013a, 2016, 2019), and were thus adopted here (see Table 1). For the Gloria Knolls Slide, slide dimensions were determined using bathymetry data containing the slide scar (Puga-Bernabéu et al. 2016, 2019). The slide is believed to have failed sequentially in multiple phases, forming what is known as a larger “slide complex”. Here, we modelled what was determined to be the worst-case scenario of these failure phases (i.e., “Event 2, Worst-Case Scenario”, see Puga-Bernabéu et al. 2019). This case was selected to represent one of the most severe submarine landslide cases for this region, as the Gloria Knolls Slide is, thus far, the largest documented slide complex (total volume ≈ 32 km³) on the northeastern Australian margin (Puga-Bernabéu, in prep). For the Noggin Block, the initial dimensions were determined from a rigorous, modelling-based slope stability analysis conducted for the block (Puga-Bernabéu et al. 2013a). This feature is comparatively small; the estimated slide volume is ~ 0.77 km³ (using the volume formulas of Enet and Grilli 2007). However, the block is relatively shallow, resting on the upper slope (~ 400 m depth). An additional sensitivity analysis was conducted to test the impact of failure depth on the initial tsunami wave height (see Sect. 4.2).

For both landslide cases, kinematic parameter a₀ was determined using the semi-empirical formulations of Enet and Grilli (2007), and the peak slide velocity was prescribed a value of 25 m/s. This peak velocity is of similar magnitude to those recorded by submarine cable breaks during the Grand Banks Event (i.e., 20–25 m/s; Fine et al., 2005). A landslide density of 2000 kg/m³ was informed by sediment core measurements obtained by Ocean Drilling Program (ODP) Leg 133 Site 819, which was drilled ~ 70 km north of the Noggin Block and the Gloria Knolls Slide (Davies et al. 1991). Each simulation was run for a landslide failure duration of 3 min at 100 m resolution horizontally and at 5 sigma layers vertically.

3.2 Tsunami propagation

The resulting ocean free surface elevations, as well as the depth-averaged zonal and meridional velocities, were smoothed and re-interpolated from the tsunami generation model outputs to set the initial conditions for the wave propagation model. Tsunami propagation was modelled using FUNWAVE-TVD (Shi et al. 2012), a widely used, fully nonlinear Boussinesq tsunami propagation code that has been validated against NOAA’s National Tsunami Mitigation Program benchmark requirements (NTHMP, 2012). The model captures wave behaviours such as shoaling, dissipation via bottom friction and wave breaking, and frequency dispersion (Shi et al. 2012).

For the earthquake scenarios, tsunami propagation was simulated across the Coral Sea using a 1 arcminute ETOPO1 grid (Amante and Eakins 2009). Smaller nested grids were used to resolve the earthquake-generated waves upon arrival to the continental shelf. These grids were generated from a 100 m resolution bathymetric dataset spanning the entire northeastern Australian margin, including the GBR (i.e. “3DGBR”, Beaman 2010; see Fig. 3a). Waves were introduced into the smaller nested grids via a one-way coupling scheme. Near-field landslide scenarios were also simulated with grids generated from the 3DGBR bathymetric dataset. Bathymetry for all cases was smoothed using a Gaussian filter to prevent numerical instability incited by steep bathymetric slopes.

The spatial resolution of the smaller model domains was carefully selected using a range of sensitivity analyses (see Online Resource 2). For the earthquake scenarios, a 200 × 200 m grid is deemed sufficient to resolve interactions between the propagating waves and the seafloor. The Gloria Knolls Slide and the Noggin Block potential failure necessitated finer resolution grids to adequately resolve shoaling and scattering processes (100 m and 50 m resolution, respectively).

It is important to note here that although Geowave also has the ability to simulate tsunami generation and propagation by both coseismic slip and landslide sources, we opted to use updated models that more explicitly resolve processes involved in landslide tsunami generation (i.e. the non-hydrostatic formulations of NHWAVE) and more accurately represent frequency dispersion of propagating gravity waves (i.e. the improved fully nonlinear, Boussinesq formulations of FUNWAVE-TVD). Dispersive effects become more critical to simulate for far-field and landslide tsunami sources (Tehranirad et al. 2015).

3.3 Run-up estimation

In the absence of the nearshore high-resolution bathymetric and topographic data (< 50 m) required to accurately resolve onshore tsunami inundation, final estimated run-up distributions were calculated using virtual tide gauges placed along the shoreline in ~ 25 m water depth d using the following equation:

$$R={A(d)}^\frac{4}{5} \cdot {d}^\frac{1}{5}$$

(1)

where R is the estimated run-up and A(d) is the maximum wave amplitude at a virtual gauge location at depth d. This formula is based on the conservation of wave energy flux and applies to both breaking and non-breaking waves (Ward and Asphaug 2003).

3.4 Testing the impact of the GBR on tsunami propagation

A major objective of this study is to test whether the structural complexity of the GBR plays a role in attenuating tsunami wave energy. The GBR exhibits structural complexity at two predominant spatial scales. Firstly, due to the morphological diversity of individual species, coral cover is structurally complex on the meter to sub-meter scale (Nelson 1996; Graham and Nash 2013). We hereafter refer to the structural complexity of coral cover as “ecosystem-scale” complexity. In a modelling context, this “ecosystem-scale” complexity cannot be resolved in the computational domain and must be parameterized (see Sect. 3.4.1). Secondly, the GBR exhibits structural complexity at the > 1 km scale. The reef structure itself is composed primarily of completely submerged or semi-submerged carbonate platforms. These features create complex positive relief on the submerged continental shelf, and much of this relief (aside from smaller, deeper pinnacles and terraces) is resolved by the 100 m resolution 3DGBR bathymetric dataset (Beaman 2010). Thus, the reef structure can be adequately resolved in the computational domain. We hereafter refer to complexity introduced by the reef structure as “bathymetric-scale” complexity.

The following sections detail how the impact of the GBR’s structural complexity at both the ecosystem scale and bathymetric scale was tested.

3.4.1 Ecosystem-scale complexity: coral cover parameterization

In FUNWAVE-TVD, bottom shear stress τ is calculated using the standard quadratic drag law (Shi et al. 2016):

$$\tau =\frac{1}{2}\rho {C}_{D}{U}^{2}$$

(2)

where C_D is the non-dimensional bottom friction coefficient, $\rho$ is the density of water, and U is the particle velocity at the seabed. A variable bottom friction coefficient was established throughout the domain, where it was altered according to the presence or absence of coral cover on reef platforms. A value of C_D = 0.1522 was prescribed to reef platforms to simulate coral cover (average depth of platforms ≈ 14.9 m). This value was obtained from a prior field investigation of the hydraulic roughness of coral reefs, which was conducted at John Brewer Reef, a reef platform within the GBR that lies close to the study region (Nelson 1996; ~80 km from the computational domain). Additionally, this coefficient falls well within the range of values obtained for other reefs (Monismith et al. 2013). All other areas of the computational domain where prescribed the conventional value of C_D = 0.0025, which is representative of sand-covered seafloor (Grilli et al. 2015). This approach was used to create the “coral cover” scenarios, where the ecosystem-scale structural complexity of the GBR was taken into account in tsunami propagation simulations (Fig. 3a).

To test the impact of coral cover on tsunami attenuation, the “coral cover” scenarios were then compared to “smooth platform” scenarios, where coral cover was effectively removed. In the “smooth platform” scenarios, all areas of the bottom boundary, reef platforms included, were prescribed a standard bottom friction coefficient value of C_D = 0.0025.

3.4.2 Bathymetric-scale complexity: testing the impact of the reef platforms

Larger-scale, bathymetric complexity is introduced by the reef structure itself, which is composed primarily of reef platforms. Testing the impact of these platforms on tsunami propagation requires artificial bathymetry, where the positive relief formed by the platforms is removed from the shelf (Fig. 3b). Platforms were removed by “cookie-cutting” the bathymetry, removing areas of the mid- to outer-shelf containing the reef platforms. The bathymetry was then linearly interpolated and smoothed over the cookie-cut areas employing a Gaussian filter. This modified bathymetry was then used in the “no reef platforms” scenarios.

3.5 Testing the additional effect of tidal phase

As the central northeastern Australian margin is a meso-tidal environment, water depths over the reef platforms can vary significantly over several hours. Consequently, tidal phase has been shown to modulate the degree of wind wave attenuation (Young and Hardy 1993). To test the impact of tidal phase on tsunami propagation, two additional scenarios were configured: one where the highest spring tide (1.75 m above MSL) and one where the lowest spring tide (1.75 m below MSL) coincided with tsunami arrival at the GBR.

4 Results

4.1 Earthquake tsunami generation and regional propagation

For the hypothetical M_w 8.0, 8.5, and 9.0 Solomon Islands earthquake scenarios, the generation model simulates initial peak wave amplitudes of 0.32 m, 1.7 m, and 9.7 m, respectively (Fig. 4). The tsunamis in each case then propagate across the Coral Sea to the outer GBR margin after an approximately 3.5 h travel time, which is consistent with previous travel times observed for the Solomon Islands source zone (NGDC/WDS 2020). Upon arrival to the outer Australian continental shelf within the nested domain, wave amplitudes range from ~ 1–2 cm for the M_w 8.0 case, ~ 6–10 cm for the M_w 8.5 case, and ~ 30–60 cm for the M_w 9.0 case.

Fig. 4

source are 0.32 m, 1.7 m, and 9.7 m, respectively. The simulated propagation time represented here is ~ 8 h to allow waves to reach all parts of the bathymetric domain

Maximum wave amplitudes simulated by FUNWAVE-TVD for the hypothetical M_w 8.0 a, M_w 8.5 b, M_w 9.0 c Solomon Islands earthquake sources. Initial maximum wave amplitudes at the

4.2 Landslide tsunami generation

The landslide generation model NHWAVE simulates an ~ 18 m-high seaward-propagating wave crest and a ~ 9 m-high landward-propagating wave crest for the Gloria Knolls Slide (Fig. 5), assuming the previously determined worst-case scenario (Puga-Bernabéu et al. 2019). For the potential collapse of the Noggin Block, the landslide generation model simulates a ~ 1.3 m-high seaward-propagating crest and a ~ 3.5 m-high landward-propagating crest (Fig. 6a). Sensitivity analyses indicate that initially generated wave amplitudes are responsive to moderate changes in depth (± 100 m). If the block was to initially fail 100 m deeper (500 m depth), the wave amplitude of the landward-propagating crest reaches ~ 2.5 m, about 71% of its original value (Fig. 6b). On the other hand, should the block fail at a 100 m-shallower depth (300 m depth), the wave amplitude peaks at ~ 4.8 m, growing roughly 37% (Fig. 6c). For the subsequent simulations of tsunami propagation, the main Noggin Block scenario (failure depth = 400 m) is implemented.

Fig. 5

Instantaneous free surface elevation at t = 9 min for the Gloria Knolls landslide tsunami scenario, simulated using NHWAVE. Wave amplitude peaks at η ≈ 18 m. The smaller peak is the landward-propagating wave, and it peaks at η ≈ 9 m

Fig. 6

Instantaneous free surface elevations at t = 9 min for the potential Noggin Block collapse, simulated using NHWAVE. The main scenario a assumes a failure depth of ~ 400 m, where the peak wave amplitude for the landward-propagating crest reaches ~ 3.5 m. A sensitivity test indicates that a 100-m-deeper failure b would result in a substantially smaller wave crest (η_max ≈ 2.5 m, 71% of its original value). A 100-m-shallower failure c would result in a larger initial wave crest (η_max ≈ 4.8 m, 37% greater than its original value)

4.3 Nearshore earthquake tsunami propagation

Results indicate that the GBR’s buffering impact on the earthquake-generated tsunami, which originates in the Solomon Islands source zone, depends on the magnitude of the initial earthquake. Turning firstly to the hypothetical M_w 8.0 earthquake scenario (Fig. 7a), maximum wave amplitudes across the domain remain under 5 cm when coral cover is present atop the reef platforms (i.e. when ecosystem-scale complexity is high), where maximum estimated run-up R_max reaches ~ 6.2 cm. When coral cover is removed (Fig. 7b), maximum wave amplitudes increase marginally or remain the same, growing 2% on average along the 25 m isobath. Estimated run-ups follow a similar trend (R_max ≈ 6.4 cm). Finally, when reef platforms are removed from bathymetry (Fig. 7c), offshore wave amplitudes increase a bit more substantially (17% on average), but still fall below ~ 5 cm across the domain. The maximum run-up estimate remains at a similar elevation (R_max ≈ 6.7 cm, Fig. 7d).

Fig. 7

Maximum wave amplitude distributions for the hypothetical M_w 8.0 Solomon Islands earthquake scenario simulated with a “coral-covered platforms” (bottom friction coefficient C_D = 0.1522 on platforms, shown in white) b “smooth platforms” (C_D = 0.0025), and c “no reef platforms”. d Corresponding maximum offshore wave amplitude and estimated run-up distributions. Maximum run-up estimates are 6.2 cm for the “coral-covered platforms” scenario, 6.4 cm for the “smooth platforms” scenario, and 6.7 cm for the “no reef platforms” scenario. Offshore wave amplitudes were interpolated along the 25 m isobath

For the hypothetical M_w 8.5 Solomon Islands earthquake scenario, the GBR, both in terms of its ecosystem-scale and bathymetric-scale complexity, appears to have slightly more of a buffering impact on offshore tsunami amplitudes and estimated run-up. When coral cover is present (Fig. 8a), wave amplitudes landward of the GBR range from ~ 5–10 cm, with an R_max estimate of ~ 26 cm. When platforms are smoothed (Fig. 8b), these amplitudes grow, increasing 7% on average along the 25 m isobath. The maximum run-up estimate also increases slightly (R_max ≈ 28 cm). Wave amplitudes similarly increase when reef platforms are removed (Fig. 8c; 13% average increase along the 25 m isobath; R_max ≈ 32 cm). Overall, the changes in the amplitude and run-up distributions are moderate for this case (Fig. 8d).

Fig. 8

Maximum wave amplitude distributions for the hypothetical M_w 8.5 Solomon Islands earthquake scenario simulated with a “coral-covered platforms” (bottom friction coefficient C_D = 0.1522 on platforms, shown in white) b “smooth platforms” (C_D = 0.0025), and c “no reef platforms”. d Corresponding maximum offshore wave amplitude and estimated run-up distributions. Maximum run-up estimates are ~ 26 cm for the “coral-covered platforms” scenario, ~ 28 cm for the “smooth platforms” scenario, and ~ 32 cm for the “no reef platforms” scenario. Offshore wave amplitudes were interpolated along the 25 m isobath. For animations of tsunami propagation for the “coral-covered platforms” and “no reef platforms” scenarios, see Online Resources 3 and 4

The GBR has a much more substantial impact on the propagating tsunami when considering the hypothetical M_w 9.0 Solomon Islands earthquake source. Overall, the M_w 9.0 earthquake-generated tsunami is significantly larger in amplitude than its smaller-magnitude counterparts. When coral cover is present on reef platforms, maximum offshore wave amplitudes range from about 0.2–0.4 m landward of the GBR (Fig. 9a), resulting in a maximum estimated run-up of ~ 0.85 m. When platforms are smoothed (Fig. 9b), amplitudes increase (18% on average along the 25 m isobath), particularly directly landward of broad reef platforms. Likewise, the maximum estimated run-up increases when platforms are smoothed, reaching 1 m. Finally, when reef platforms are removed from bathymetry (Fig. 9c), amplitudes increase substantially on the shelf (51% on average along the 25 m isobath), leading to a maximum estimated run-up of ~ 1.2 m (Fig. 9d).

Fig. 9

Maximum wave amplitude distributions for the hypothetical M_w 9.0 Solomon Islands earthquake scenario simulated with a “coral-covered platforms” (bottom friction coefficient C_D = 0.1522 on platforms, shown in white) b “smooth platforms” (C_D = 0.0025), and c “no reef platforms”. d Corresponding maximum offshore wave amplitude and estimated run-up distributions. Maximum run-up estimates are 0.85 m for the “coral-covered platforms” scenario, 1.0 m for the “smooth platforms” scenario, and 1.2 m for the “no reef platforms” scenario. Offshore wave amplitudes were interpolated along the 25 m isobath

Figure 10 shows the percentage increase exhibited by both offshore wave amplitude and predicted run-up when both the ecosystem-scale and bathymetric-scale complexity of the GBR is removed. This gives an indication of the relative degree to which the GBR attenuates tsunami wave energy. Firstly, considering ecosystem-scale complexity in isolation, when coral cover is removed, under the M_w 8.0 scenario (Fig. 10a), wave amplitudes are slightly larger on a percentage-wise basis compared to when coral cover is present, ranging from 0–4% increase within the study area. For the M_w 8.5 scenario, this percentage increase heightens, ranging from 1–15%. Finally, for the largest earthquake scenario (M_w 9.0), amplitudes increase substantially, ranging from 3–31% higher compared to when coral cover is present. Percentage increases in the estimated run-up distributions follow similar patterns. Amplitude and run-up increases are highly variable alongshore, with the largest peaks occurring directly behind shelf areas with broad, shallow reef platforms. For instance, the city of Cairns (latitude ≈ 16.8°S) seems to benefit from being situated behind a wide, shallow reef platform that lies in the path of the tsunami. The overall trend indicates that the attenuating effect of coral cover increases with the magnitude of the earthquake source.

Fig. 10

Percentage increases in both earthquake tsunami amplitude and estimated run-up when a coral cover is removed and b reef platforms are removed. Amplitudes, which were also used to calculate run-up, were extracted along the 25 m isobath

The second panel of Fig. 10 reflects the very substantial impact of ecosystem-scale and bathymetric-scale complexity (i.e. coral cover and reef platforms). When coral cover and reef platforms are removed, wave amplitudes and run-ups increase considerably for the M_w 8.0 scenario (range 0–48%). Notably, at a few locations, this percentage dips marginally below zero (-5% maximum), indicating that these areas would experience a decrease in offshore amplitudes and estimated run-ups if reef platforms were not present. Amplitude and run-up distributions follow a similar pattern, increasing overall for the M_w 8.5 (range 7–70%), and again for the M_w 9.0 (range 20–90%). These results reflect the significant combined attenuative impact of both coral cover and the reef platforms on the propagating tsunamis, an impact which increases with earthquake source magnitude. We again note the immense variability of the amplitude and run-up increases alongshore for each earthquake scenario.

For all cases, the first tsunami waves arrive at the coast after an approximately 4 h travel time from the Solomon Islands source zone. When passing over the shelf, the tsunami experiences refraction, diffraction, shoaling, and focusing. In particular, broad, moderately deep platforms tend to focus tsunami wave energy towards shore (e.g., Fig. 9b). When platforms are removed, this behaviour disappears. For animations of the M_w 8.5 scenario simulated with coral cover and no reef platforms, see Online Resources 3 and 4.

4.4 Nearshore landslide tsunami propagation

Simulations indicate that the impact of ecosystem-scale and bathymetric-scale complexity on tsunami attenuation is sizeable for the landslide-generated cases considered on this margin. Turning firstly to the previously termed “worst-case scenario” for the Gloria Knolls Slide (Puga-Bernabéu et al. 2016), when reef platforms are covered by coral, offshore amplitudes markedly decline from over ~ 4 m to under ~ 2 m landward of the platforms (Fig. 11a), and maximum estimated run-up reaches up to ~ 2.2 m. When coral cover is removed (Fig. 11b), offshore amplitudes along the 25 m isobath nearly double, increasing by a factor of ~ 1.9 on average. Maximum estimated run-up rises to ~ 3.9 m under the “smooth platforms” simulation. When reef platforms are removed (Fig. 11c), offshore amplitudes more than quadruple on average (fold change: ~ 4.6), when compared to the “coral-covered platforms” scenario. When platforms are absent, estimated maximum run-up increases again, reaching 4.6 m (Fig. 11d). The total elapsed time between tsunami generation and the arrival of the first waves is ~ 1.5 h.

Fig. 11

Maximum wave amplitude distributions for the Gloria Knolls Slide (worst-case scenario) simulated with a “coral-covered platforms” (bottom friction coefficient C_D = 0.1522 on platforms, shown in white) b “smooth platforms” (C_D = 0.0025), and c “no reef platforms”. d Corresponding maximum offshore wave amplitude and estimated run-up distributions. Maximum run-up estimates are ~ 2.2 m for the “coral-covered platforms” scenario, ~ 3.9 m for the “smooth platforms” scenario, and ~ 4.6 m for the “no reef platforms” scenario. Offshore wave amplitudes were interpolated along the 25 m isobath. For animations of the “coral-covered platforms” and “no reef platforms” scenarios, see Online Resources 5 and 6

Results for the Noggin Block potential slide are very similar to that of the Gloria Knolls Slide, though it produces a smaller tsunami (see Sect. 4.2). Assuming healthy reef growth (Fig. 12a), offshore amplitudes remain under ~ 1 m, where they sharply decline upon passing over the GBR platforms. Maximum estimated run-up for this scenario is ~ 1.4 m. When coral cover is removed (Fig. 12b), offshore amplitudes along the 25 m isobath increase by a factor of ~ 2 on average, with the maximum run-up rising to 1.8 m. Finally, when platforms are removed from the simulations (Fig. 12c), offshore amplitudes along the 25 m isobath are, on average, 4.5 times larger than the original “coral cover” scenario. Peak estimated run-up reaches ~ 2.8 m under the “no reef platforms” scenario (Fig. 12d). The total time between tsunami generation and the arrival of the first waves is similar to the Gloria Knolls landslide tsunami (~ 1.5 h).

Fig. 12

Maximum wave amplitude distributions for the Noggin Block potential landslide scenario simulated with a “coral-covered platforms” (bottom friction coefficient C_D = 0.1522 on platforms, shown in white) b “smooth platforms” (C_D = 0.0025), and c “no reef platforms”. d Corresponding maximum offshore wave amplitude and estimated run-up distributions. Maximum run-up estimates are ~ 1.4 m for the “coral-covered platforms” scenario, ~ 1.8 m for the “smooth platforms” scenario, and ~ 2.9 m for the “no reef platforms” scenario. Offshore wave amplitudes were interpolated along the 25 m isobath. For animations of the “coral-covered platforms” and “no reef platforms” scenarios, see Online Resources 7 and 8

Figure 13 shows the overall change in offshore wave amplitude and estimated run-up when coral cover and reef platforms are removed from simulations, this time represented in terms of fold change rather than percentage change. For each landslide case, offshore amplitudes along the 25 m isobath, along with estimated run-ups, tend to double when coral cover is removed (Fig. 13a). When platforms are removed, the amplitudes and run-ups increase significantly for each case, but more so for the Noggin Block potential slide (Fig. 13b). Again, we highlight the enormous along-shore variability in amplitude and run-up change across simulations.

Fig. 13

Fold-change increase in both landslide tsunami amplitude and estimated run-up when a coral cover is removed and b reef platforms are removed. Amplitudes, which were also used to calculate run-up, were extracted along the 25 m isobath

Landslide tsunamis across both cases exhibit common behaviours. Amplitude and run-up distributions follow a localized bell-curve due to radial damping, a standard process undergone by point-source tsunamis (Brune et al. 2010; Harbitz et al. 2006). Additionally, reef platforms greatly interfere with these comparably shorter waves as they traverse the shallow continental shelf (Harbitz et al. 2006). For animations of both the Gloria Knolls slide scenario simulated with coral-covered platforms and no reef platforms, see Online Resources 5 and 6. For the same corresponding Noggin Block landslide tsunamis, see Online Resources 7 and 8.

4.5 Tidal impacts on tsunami propagation

The additional impact of tide level was tested for the M_w 8.5 Solomon Islands earthquake scenario, the Gloria Knolls Slide scenario, and the Noggin Block potential slide scenario. Results indicate a minimal impact of tide level on the degree of attenuation of the M_w 8.5 earthquake-triggered tsunami (Fig. 14a), where amplitudes were 1.6% lower on average at low spring tide (1.75 m below MSL; Fig. 14b) and 2.6% higher on average at high spring tide (1.75 m above MSL; Fig. 14c). Offshore amplitude and run-up distributions along the 25 m isobath are very similar for all tide cases (Fig. 14d). For the Gloria Knolls Slide (Fig. 15a), the effect of tides is more pronounced, where amplitudes decrease 11% on average during low spring tide (Fig. 15b) and increase 17% on average at high spring tide (Fig. 15c). Similarly, for the Noggin Block potential slide scenario (Fig. 16a), amplitudes were 16% lower on average at low spring tide (Fig. 16b) and 6% higher on average at high spring tide (Fig. 16c).

 Fig. 14

Maximum wave amplitude distributions for the hypothetical M_w 8.5 Solomon Islands earthquake scenario simulated at a mean sea level (MSL, bottom friction coefficient C_D = 0.1522 on platforms, shown in white) b low spring tide (1.75 m below MSL, C_D = 0.1522 on platforms), and c high spring tide (1.75 m above MSL, C_D = 0.1522 on platforms). d Corresponding maximum offshore wave amplitude and estimated run-up distributions. Maximum run-up estimates are 26 cm for the MSL scenario, 25 cm for the low spring tide scenario, and 26 cm for the high spring tide scenario. Offshore wave amplitudes were interpolated along the 25 m isobath

 Fig. 15

Maximum wave amplitude distributions for the Gloria Knolls Slide scenario simulated at a mean sea level (bottom friction coefficient C_D = 0.1522 on platforms, shown in white) b low spring tide (1.75 m below MSL, C_D = 0.1522 on platforms), and c high spring tide (1.75 m above MSL, C_D = 0.1522 on platforms). d Corresponding maximum offshore wave amplitude and estimated run-up distributions. Maximum run-up estimates are 1.4 m for the MSL scenario, 1.4 m for the low spring tide scenario, and 1.3 m for the high spring tide scenario. Offshore wave amplitudes were interpolated along the 25 m isobath

 Fig. 16

Maximum wave amplitude distributions for the Noggin Block potential slide scenario simulated at a mean sea level (bottom friction coefficient C_D = 0.1522 on platforms, shown in white) b low spring tide (1.75 m below MSL, C_D = 0.1522 on platforms), and c high spring tide (1.75 m above MSL, C_D = 0.1522 on platforms). d Corresponding maximum offshore wave amplitude and estimated run-up distributions. Maximum run-up estimates are 1.4 m for the MSL scenario, 1.4 m for the low spring tide scenario, and 1.3 m for the high spring tide scenario. Offshore wave amplitudes were interpolated along the 25 m isobath

5 Discussion

5.1 The impact of the GBR’s ecosystem-scale complexity on tsunami propagation

Our results show that tsunamis are strongly impacted by the presence of coral cover in the GBR. Across many of the “coral-covered platforms” simulations, maps showing maximum wave amplitude distributions show clear “shadow zones” landward of reef platforms, where amplitudes markedly decrease. These impacts are especially pronounced for the M_W 9.0 Solomon Islands earthquake scenario (Fig. 9), the Gloria Knolls submarine landslide scenario (Fig. 11) and the Noggin Block potential submarine landslide scenario (Fig. 12). These declines in wave amplitude are driven by elevated frictional dissipation over coral-covered reef platforms. We eliminate the possibility that wave breaking contributed to energy dissipation, as wave breaking was not detected in any of the simulations due to the tsunamis’ large wavelengths in comparison to their amplitudes. These results reaffirm the prevailing notion that the GBR acts as a regional buffer to tsunamis (Baba et al. 2008; Wei et al. 2015; Xing et al. 2015; Webster et al. 2016; Puga-Bernabéu et al. 2019). They are also consistent with previous findings from other modelling studies, especially those that include wider reef platforms in their assessments (Kunkel et al. 2006; Gelfenbaum et al. 2011; Yao et al. 2012), which allows the cumulative impact of frictional dissipation to dominate. Therefore, we propose that the effect of live coral cover should be directly incorporated into future hazard assessments of the northeastern Australian margin, as we anticipate it will have a buffering impact on propagating tsunamis.

The energy-diminishing impact of coral cover becomes most apparent when comparing the “coral-covered platforms” simulations with the “smooth platforms” simulations. When coral cover is removed, amplitudes increase across each source scenario tested here. Notably, run-up projections increase as much as 24% for the M_w 9.0 earthquake source (Fig. 10), and they exhibit a maximum of a nearly fourfold change for the Noggin Block potential slide (Fig. 13). These increases in amplitude and run-up imply that while coral cover in the GBR may currently have a buffering effect on tsunami wave energy, this effect may diminish as reef ecosystems in the GBR continue to decline under the physiological stressors (e.g., heat stress, acidity stress) that accompany anthropogenic climate change (Hughes et al. 2018). Generally speaking, the structural complexity of coral reefs is expected to deteriorate as reef-building species are lost and as ecosystems transition to algal-dominated states (Bellwood et al. 2004; Alvarez-Filip et al. 2009; Wild et al. 2011). This deterioration of structural complexity is expected to lessen frictional dissipation of wind wave energy (Harris et al. 2018; de Lalouvière et al. 2020). Based on our results, we expect a similar outlook for tsunami wave hazards. This loss of buffering capacity may be further compounded by the effects of sea level rise, where some assessments have forecasted heightened tsunami hazard under current projections (Li et al. 2018; Nagai et al. 2020).

Across source scenarios, there are prominent discrepancies in the magnitude of the amplitude and run-up increases when coral cover is removed. For instance, while the M_w 8.0 earthquake scenario experiences marginal increases (4% maximum, see Fig. 10), the M_w 9.0 scenario experiences substantial jumps in offshore amplitude (up to 31%) when platforms are smoothed. This implies that the degree of coral-induced frictional dissipation at bed is different across source scenarios. Our findings demonstrate that these differences in frictional dissipation are directly related to wave amplitude (and thus, wave energy). Particle velocity (note: this is different to wave celerity) is a function of wave amplitude (Nielsen 1992), and therefore, waves of differing amplitudes experience different degrees of dissipation due to shear stress at bed (Fig. 17). This amplitude-mediated discrepancy in particle velocity is best exemplified by comparing earthquake scenarios, where tsunami amplitude was altered by changing the magnitude and slip displacement of the initial coseismic source (Fig. 4, see Table 1 for source parameters). For the M_w 8.0 Solomon Islands earthquake scenario, bed particle velocities are relatively low (< 1 cm/s) throughout the computational domain given the relatively low tsunami amplitudes produced by the source. However, for the M_w 8.5 and M_w 9.0 earthquake scenarios, particle velocities are much higher on the shelf (> 5 cm/s). Moreover, in their corresponding “smooth platforms” simulations, particle velocities are more elevated atop the reef platforms than in the “coral-covered platforms” simulations, which further reflects the dissipative effect of coral cover. As wave energy dissipation through shear stress is proportional to the square of the particle velocity (see Eq. 1), the higher velocities computed for higher-magnitude earthquake tsunamis result in greater overall wave energy dissipation via bottom friction when coral cover is present. This also explains why a relatively large degree of attenuation is observed for the landslide-generated tsunamis, both of which produce similarly high waves (9 m and 3.5 m for the landward-propagating waves, respectively). Our results show that tsunami amplitude, which ultimately depends on the magnitude and proximity of the triggering source, should also be considered when examining the buffering capacity of natural defences such as coral reefs (Fig. 17).

Fig. 17

Maximum bed particle velocities (in cm/s) across the computational domain for each the M_w 8.0 Solomon Islands earthquake scenario (top row), the M_w 8.5 earthquake scenario (middle row), and the M_w 9.0 earthquake scenario (bottom row). Columns are aligned based on their corresponding “coral-covered platforms” simulations (left column), “smooth platforms” simulations (middle column), and “no reef platforms” simulations (right column)

While the GBR generally acts as a buffer to tsunami wave energy, despite its namesake, the GBR itself does not form a continuous barrier on the mid- to outer shelf, especially in the central region (Fig. 3). As a result, the buffering effect offered by coral cover varies considerably alongshore. Turning again to the Solomon Islands earthquake scenarios (Fig. 10), when coral cover is removed, the largest increases in wave amplitude and run-up tend to occur landward of broad reef platforms (see also Fig. 9a, b). On the other hand, areas that lie between inter-reef passages, or gaps, exhibit smaller increases in amplitude and run-up. This phenomenon is consistent across source scenarios, and it is particularly pronounced in cases where tsunami amplitudes are relatively high. This implies that the protectiveness offered by coral cover varies alongshore because of platform placement; if coral-covered platforms (particularly broad platforms) are positioned between the incoming tsunami and the shoreline, they are more inclined to dampen the tsunami.

To summarise, reef cover contributes substantially to the overall buffering capacity of the GBR, which is consistent with previous findings (e.g., Kunkel et al. 2006). However, the GBR’s buffering capacity for any given location alongshore depends on various site-specific factors, including the presence of coral cover, the relative positioning of the platforms, and tsunami amplitude.

5.2 The impact of the GBR’s bathymetric-scale complexity on tsunami propagation

Our simulations reveal the remarkably complex ways in which tsunami waves interact with the larger-scale bathymetric features (i.e., platforms, shoals, etc.) that comprise the GBR. Of particular note is the platforms’ ability to focus tsunami wave energy towards shore (see Fig. 8 and Online Resource 3 for the M_w 8.5 Solomon Islands earthquake simulations). In a manner analogous to a convex lens focussing light, platforms cause the incoming tsunami waves to refract inwards towards their shallower depths, inciting shoaling, positive wave interference, and subsequent heightening of wave trains. Shoaling and heightening of tsunami waves over shallow reefs has been observed by others, both from field-based and modelling evidence (Chatenoux and Peduzzi 2005; Gelfenbaum et al. 2011). Interestingly, frictional dissipation by coral cover appears to fully or partially counteract these focussing effects, where waves subsequently dampen after growing in amplitude over the platforms (e.g., Fig. 9). Consequently, removing coral cover tends to enhance the platforms’ ability to focus wave energy. This is demonstrated by the higher-amplitude, landward wave trains shown in wave amplitude distributions (e.g., Fig. 9). Some platforms appear to more effectively focus wave energy than others, and we suspect this is due to factors such as reef morphology, size, and submergence depth. A more systematic investigation of platform characteristics is warranted to test this hypothesis, particularly as coral reef cover is expected to decline in the future.

In addition to focussing effects, simulated tsunamis exhibit a complex interplay of additional behaviours when interacting with platforms, such as diffraction, reflection, and scattering of wave trains (see Online Resource 5 for example). These effects are most pronounced for the landslide-generated tsunami cases (see Sect. 4.4), and we tentatively suggest that this is due to their shorter wavelengths. Our simulations further reinforce the important role that local bathymetry plays in modulating tsunami behaviour, particularly in shallow reef environments (Baba et al. 2008; Dilmen et al. 2018). This potent, complex, and site-specific control on tsunami propagation further underscores the need to evaluate tsunami hazard on a case-by-case basis.

We also highlight the intriguing role of inter-reef passages, or gaps, in modulating tsunami behaviour as waves cross the shelf. Many have hypothesized that gaps in the reef structure worsen the tsunami hazard, as the gaps act as low-resistance conduits that amplify wave energy (Nott 1997; Liu et al. 2005; Gelfenbaum et al. 2011; McAdoo et al. 2011; Roger et al. 2014). In our simulations, porous gaps in the reef structure certainly permit wave energy to pass through to the coastline. However, there is little evidence to support the notion that the gaps amplify waves. In fact, due to focussing, amplification of wave amplitudes occurs over the platforms rather than between them (e.g., Figs. 9, 11). In the case of the GBR, many of the platforms appear to be wide enough, deep enough, far enough apart, and far enough from the coastline such that the inter-reef gaps do not pose a significant hazard. This is in contrast to many fringing reef systems, where gaps can be quite narrow, shallow, and close to shore. We therefore suggest that for the GBR, the wave focussing ability of platforms may be of greater concern for the northeastern Australian coastline than the presence of gaps in the reef structure.

Overall, the GBR’s underlying bathymetric structure contributes significantly to its buffering capacity, and this becomes apparent when platforms are removed from simulations (see Figs. 10 and 13). When platforms are removed, waves are permitted to propagate smoothly and uninterruptedly across the shelf, highlighting the highly obstructive nature of the platforms themselves. Offshore wave amplitudes and run-up distributions increase alongshore across all source scenarios when platforms are removed. These findings are consistent with previous work which suggests that bathymetric irregularities on the shelf exert large control on the eventual run-up distribution at the coast (Baba et al. 2008; Schambach et al. 2018). Even as the GBR is interrupted by gaps, the presence of the reef structure appears to provide at least some benefit to nearly all areas of the coastline examined in this study.

5.3 Broader implications surrounding the GBR’s impact on tsunami hazard

This study has revealed wider implications for communities situated along the northeastern Australian coastline. Firstly, from a mitigation perspective, the GBR may offer greater protection for more severe tsunami events. In particular, the GBR may reduce the hazard for near-field landslide sources, which are notoriously difficult to predict and forecast (Tappin et al. 1999; Harbitz et al. 2014). While this may take some pressure off early warning systems, we stress that coastal communities should not rely upon the GBR alone to reduce their vulnerability to tsunami hazards. A holistic strategy for tsunami hazard preparedness ultimately should include risk awareness, hazard education, resilient infrastructure, and robust early warning systems (Baird et al. 2005; Liu et al. 2005; Dominey-Howes et al. 2007; Mori et al. 2011).

Secondly, from a future mitigation perspective, our work suggests that declining coral health, which is associated with globally mediated anthropogenic climate change (De’ath et al. 2012; Hughes et al. 2018), will have an overall adverse effect on the GBR’s defensive capability. In this context, today’s reef-buffering asset may be tomorrow’s liability. Areas of shoreline that are best-protected by broad, expansive coral-covered platforms may experience the highest inundation risk in the future as coral die-off continues and as architectural complexity deteriorates (Alvarez-Filip et al. 2009), enhancing the platforms’ ability to focus energy towards shore rather than attenuating it. These local differences reinforce the need for site-specific hazard assessments when considering tsunami hazard on the northeastern Australian margin in the future.

5.4 Reconciling differing interpretations of coral reef impact on tsunamis

In light of our results, we address some of the contrasting interpretations in the literature around the impact of coral reefs on tsunami hazards. Firstly, while the GBR, being an offshore barrier system, buffers the tsunami hazard for the more distant Australian coastline, other reef environments (in particular, narrow fringing reefs that surround populated inner islands) could exacerbate tsunami hazard through behaviours such as shoaling, focussing, and bore formation (Chatenoux and Peduzzi 2005; Fritz et al. 2011; Gelfenbaum et al. 2011; Yao et al. 2012). Indeed, our simulations showcase shoaling and focussing on platforms, which locally augment wave amplitudes at the intra-platform scale. A more rigorous inundation study would be needed to confirm whether this translates to increased hazard within the lagoons, shoals, and islands that rest within the platforms. Therefore, coral reefs could have either beneficial or detrimental effects on the overall hazard depending on the type of reef system in question and the proximity of coastal communities and assets to the site of the most severe shoaling/focussing. In the debate surrounding reef protectiveness against tsunamis, a distinction must be made between fringing reef systems and offshore barrier systems, as they have different implications for proximity to wave focussing effects, and therefore, exposure.

On the other hand, we also note potential ambiguities around the ways in which the impact of coral reefs is reported in post-tsunami field surveys. From our simulations and others (Kunkel et al. 2006; Uslu et al. 2010; Gelfenbaum et al. 2011), there is evidently a strong theoretical basis to support the fact that coral reefs can dissipate tsunami wave energy, reducing the tsunami hazard. However, this overall reduction in hazard may not be sufficient to completely reduce the physical vulnerability and exposure of coastal communities (Uslu et al. 2010). When discussing the buffering role of reefs, many have highlighted that despite being within close proximity to reefs, coastal assets have nonetheless been destroyed during tsunami events (e.g., Baird et al. 2005), leading some to conclude that coral reefs provide no protective benefit to coastal communities. In these cases, the reefs could very well have buffered the overall tsunami hazard, reducing the overall inundation and run-up extent. However, this protective benefit may not have been sufficient to completely shield coastal communities that were situated close to shore. Care must be taken when retrospectively interpreting the role that coral reefs may have played in reducing tsunami hazard along a shoreline, and a clear distinction should be made between hazard reduction and risk reduction, which lies at the intersection between hazard, exposure, and vulnerability.

5.5 Study limitations and future work

Uncertainties persist that could complicate such future tsunami hazard assessments in coral reef environments. Firstly, at the ecosystem scale, the relationship between coral rugosity and community composition requires more precise quantification on an intra-reef platform scale (Rogers et al. 2016). This will continue to be a pressing task in the future, as ecological shifts may be precipitated by both the immediate aftermath of the tsunami impact and longer-term environmental changes, thus affecting ecosystem-scale structural complexity (Madin and Connolly 2006; Alvarez-Filip et al. 2009; Ferrari et al. 2016; Hughes et al. 2018). While platform degradation and bioerosion is largely anticipated to flatten coral reefs (Alvarez-Filip et al. 2009), the shorter-term impact of these and other stressors on ecosystem-scale rugosity is still not precisely known. These ecosystems should be stringently monitored to better assess how coastal hazard severity as a whole will be transformed in these areas. Additionally, the approach used to parameterize bottom shear stress, though very common both in the field and in modelling studies, may need to be reconfigured to account for more complex tsunami interactions and subgrid turbulent dissipation within the 3D reef structure (Lowe et al. 2008; Kim et al. 2009; Rosman and Hench 2011). Moreover, these more complex interactions may be better represented by a non-hydrostatic, multi-layer Navier–Stokes model rather than a depth-averaged wave model (Kazolea et al. 2019).

On a larger scale, it is worth exploring the potential impact of undular bores that could arise and break on the platforms themselves, as they could play an additional role in dissipating wave energy offshore (Grilli et al. 2012; Glimsdal et al. 2013). These wave features would not have been resolved in our coarser-resolution runs (Schambach et al. 2018), as capturing them quickly becomes very demanding computationally (< 10 m resolution required, Grilli et al. 2012). Also, while our study emphasizes the effect of the GBR on offshore amplitudes and projected run-up distributions, ultimately, tsunami-induced surges and bores deliver the force and high water levels that cause destruction to coastal communities onshore (Koshimura et al. 2009; Nistor et al. 2009; Nouri et al. 2010). Further work is warranted to establish whether the reduction in offshore wave amplitude translates to a reduced hazard onshore, and this would necessitate the deployment of higher-resolution inundation simulations.

Finally, our study was not designed to provide a reappraised, comprehensive hazard assessment for the northeastern Australian coastline although our findings suggest that the reef’s role should be considered in future assessments. That being said, we stress the need for a robust parameterization of reef roughness (Nelson 1996; Rosman and Hench 2011). Furthermore, as indicated by sensitivity analyses (see Online Resource 2), these propagation simulations require high spatial resolution (200 m for earthquake sources and 100 m or less for landslide sources) in order to properly capture the reef structure and to resolve complex tsunami-reef interactions. While this increases computational demand, we nonetheless deem it worthwhile to consider the role of the reef, as current assessments may be over-estimating tsunami risk in northeast Australia. Additionally, a more meaningful assessment of the submarine landslide tsunami hazard is needed to better understand the timing, frequency, and magnitude of these events. In the future, it may be worth considering more complex failure dynamics (i.e. landslide deformation and two-way coupling with the water column), which could alter the run-up results (Masson et al. 2006; Geist et al. 2009; Abadie et al. 2010), but we anticipate that accounting for these dynamics will not alter the overall conclusions established here about the buffering effect of the GBR. Addressing these limitations will enable more reliable forecasting as the fate of the world’s coral reefs becomes clearer with time.

6 Conclusions

This study demonstrates the nuanced interactions between tsunamis and coral reef systems. In agreement with previous work we find that the Great Barrier Reef (GBR), both in terms of coral cover and larger-scale bathymetric complexity, acts as a large-scale regional buffer against tsunamis. However, the reef appears to provide greater protection against higher-amplitude tsunamis due to the larger computed particle velocities at bed, which directly dictates the degree of frictional dissipation through shear stress. Additionally, we find that the protectiveness offered by the GBR locally depends on coral cover and platform orientation and distribution. We also find that wave focussing by reef platforms could pose a greater hazard than the gaps between platforms, which have been previously thought to amplify waves. In the context of the larger debate about whether coral reefs reduce tsunami hazards for coastal communities, we note that differing interpretations can be reconciled when considering site-specific factors.